The ICTD Government Revenue Dataset Wilson Prichard Interna1onal Center for Tax and Development
Jun 02, 2015
The ICTD Government Revenue Dataset
Wilson Prichard Interna1onal Center for Tax and Development
Overview • The ICTD GRD responds to major limits of exis1ng sources for conduc1ng
cross-‐country tax research, with major improvements in data coverage and accuracy by combining data from mul1ple sources – including a more consistent approach to natural resource revenues
• This is a cri1cal complement to work at interna1onal organiza1on to improve data over the long-‐term, as it offers a much improved founda1on for immediate research.
• However, it is a very par1al solu1on: There are inescapable limita1ons, which reflect the limits of any available sources, and the imperfec1ons of merging data from mul1ple sources
• There is a need for coopera1on and consensus to maintain the dataset as a resource for researchers while new efforts at the IMF, OECD and elsewhere begin to bear fruit.
Outline
1. Mo1va1on
2. Construc1on of the Dataset
3. Limita1ons
4. Lessons and Next Steps
Motivation • Weaknesses of exis,ng data raise serious concerns
about the robustness of tax and development research, and reduces value of data for broader descrip1ve and compara1ve exercises
• Exis1ng interna1onal sources all suffer from substan1al limita1ons – reflected in researchers relying increasingly on composite and ad hoc datasets
• However, ad hoc datasets subject to errors, lack of transparency and difficul1es of comparability
Weaknesses of Existing Sources • Missing data in sources with full country coverage • Limited coverage and comparability of regional
sources • Non-‐tax revenue o>en not included, thus giving
incomplete picture of government finances • Failure to consistently dis,nguish natural resource
revenues in most exis1ng databases • Incomplete range of revenue categories in many
researcher databases • Simple errors, most notably in researcher databases
– and oYen driven by merging of sources • Problems with inconsistencies in many GDP series
Potential for Complementarity
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
1990 1993 1996 1999 2002 2005 2008 Year
Ghana: Total tax as % GDP
K & M
GFS
IMF CR
WB WDI
AEO
Construction of the ICTD GRD 1. A Standard Revenue Classifica1on 2. Compiling Available Interna1onal Sources 3. Compiling and Adding Ar1cle IV data 4. Dealing with Natural Resources 5. Addi1onal Issues 6. A Common GDP Series 7. Manual Data Cleaning
Construction of the ICTD GRD: A Standard Revenue Classification
• Tax and Non-‐tax
• Natural Resources
• Social Contribu1ons
!!
Total!Gov’t!
Revenue!
Total!Gov’t!Revenue!Excluding!Grants!
Grants!
Tax!Revenue!
Non7Tax!Revenue!!
!
Social!Contributions!
Non7Resource!
Direct!Taxes!
Indirect!Taxes!
Non_Resource!Taxes!on!
Incomes,!Profits!and!Capital!Gains!
Property!Taxes!
Taxes!on!Individuals!
Non_Resource!Taxes!on!
Corporations!
Taxes!on!International!
Trade!
Taxes!on!Goods!and!Services!
Other!Taxes!
Sales!Taxes/VAT!
Excises!
Imports!
Exports!
Resource!Non7Tax!
!
Non7Resource!Non7Tax!Revenue!
Non7Resource!Tax!
Revenue!
Resource!Tax!
Revenue!
Resource!Direct!Taxes!
Resource!Taxes!on!Incomes,!Profits!and!Capital!Gains!
Resource!Taxes!on!Corporations!
Construction of the ICTD GRD: Compiling Alternative Sources
• IMF GFS (pre and post-‐1990) • OECD • CEPALSTAT • OECD LatAm • OECD AEO • World Bank • Keen and Mansour
Construction of the ICTD GRD; Article IV Data
• Ar1cle IV data oYen available where other sources missing – though is less rigorously reviewed, so should be used when it matches surrounding sources
• Requires careful categoriza1on, as revenue categories vary across countries and over 1me
0%
5%
10%
15%
20%
25%
30%
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Rat
io (%
)
Year
Albania: tax/GDP ratio (%)
Art. IV (GG) GFS (GG) GFS (CG) GFS (CG+SS) Michigan Ross WTD WB
Construction of the ICTD GRD: Natural Resources
Angola 1996
• Interna1onal sources are inconsistent in classifying resource revenue between taxes and non-‐tax revenue
• Non-‐resource tax revenue is the analy1cally interes1ng category, which requires excluding natural resource component of tax
• Some1mes possible using OECD, most oYen rely on IMF Ar1cle IV
Total Revenue
Total Tax Taxes on Income
Total Non-‐Tax Rev
Resource Revenue
Non-‐Resource Non-‐Tax
Pre-‐Adjustment
48.9% 48.6% 32% 0.3% -‐ 0.3%
Post-‐Adjustment
48.9% 4.8% 0.9% 44.1% 43.8% 0.3%
Construction of the ICTD GRD: Other Issues
• Consistent approach to social contribu,ons: Varia1on across sources in the inclusion of social contribu1ons can lead to incompa1bility. We report all figures inclusive and exclusive of social.
• Dealing with federal states: Focusing exclusively on central government can vastly understate tax collec1on in federal states. We adopt general government data where it is significantly different from central data, and consistent over 1me.
• Direct and Indirect Taxes: Owing to differences across sources in sub-‐categories of taxa1on, we calculate direct and indirect taxes for all country-‐years.
Construction of the ICTD GRD: Common GDP Series
• There are simple differences across sources in GDP figures, making transparency and consistency about GDP figures as important as the tax data
• Growing recogni1on that underes1ma1on of GDP can lead to vast overes1ma1on of key variables as shares of GDP
• Equally, irregular rebasing exercises can lead to major breaks in 1me series data unless applied retroac1ve to earlier periods – which is frequently not the case
Construction of the ICTD GRD: Common GDP Series
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
1980 1984 1988 1992 1996 2000 2004 2008 Year
Ghana: Total tax as % source-specific GDP
K & M GFS IMF CR WB WDI
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
12.0%
14.0%
16.0%
1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 Year
Ghana: Total tax as % common GDP series
K & M GFS IMF CR WB WDI
Construction of the ICTD GRD: Manual Data Cleaning
• Data merging would, ideally, be automated, filling gaps in a “baseline” source with pre-‐determined “second best” data
• However, inconsistencies across sources – some reflec1ng differing methods, others reflec1ng simple data discrepancies – imply that automated processes result in incompa1ble and inconsistent 1me series
• As such, it is necessary to manually clean the data to ensure consistency within countries between data sources
Developing Country Data Coverage
Data coverage is drama1cally more complete than for other sources, including the most widely used composite dataset, from the IMF FAD. ICTD GRD IMF FAD IMF Art IV IMF GFS WDI
Total Revenue 2317 1913 1484 1391 1060
Total Tax 2348 1976 1895 1396 1060
Taxes on Income, Profits and Capital Gains
1900 1909 1341 1395 1060
Taxes on Goods and Services
1952 1856 1092 1395 1060
Continued Limitations 1. S,ll significant missing data
2. Challenges in dealing with resource revenues 1. Some1mes data is not available, so countries excluded
from analysis 2. OYen impossible to dis1nguish resource revenue from
other non-‐tax revenue 3. Defini1onal issues in deciding what classifies as resource
revenue 3. Varia,on across sources o>en inexplicable, data
inherently imperfect – and merging choices inevitably subjec1ve
Lessons and Next Steps 1. Dealing with resource revenues is cri1cal, needs to be integrated with
interna1onal databases and requires a common framework
2. Ajen1on to GDP figures equallly cri1cal, and any dataset should separately provide LCU figures, % of GDP figures and clearly documented GDP series
3. Any dataset should deal with both tax and non-‐tax revenues in order to be analy1cally useful for research, while also adop1ng a consistent approach to social contribu1ons
4. There remain opportuni1es for much improved interna1onal coopera1on, as there is currently major overlap and duplica1on – some1mes even within organiza1ons – and new ini1a1ves have tended to address some, but not all, of the challenges noted here.
5. Merging data from mul1ple sources for research is fraught with risks – and is extremely 1me intensive – thus placing a premium on establishing a single accepted source, transparency and providing resources for long-‐term maintenance of the dataset